Prediction Of Virus-Receptor Interaction Based On Multiple Data Fusion

The receptor-binding is the first step of viral infection.To more effectively treat viral infectious diseases,potential virus-receptor interactions must be identified.However,considering the fact that biological data has diversity and missing values,the virus-receptor interaction prediction is still a big challenge.In this paper,we collect experimentally verified virus-receptor interactions from existing literature and related databases to expand viral Receptor.After integrating multiple data,four computational methods have been proposed based on machine learning,matrix completion and link prediction methods.These details are as follows:(1)Experimentally verified virus-receptor interactions have been collected from viral Receptor,existing literature,Viral Zone and Uni Prot KB/Swiss Prot to build an extended dataset,virtual Receptor＿sup.This dataset provides a data foundation for the prediction of virus-receptor interactions.(2)To reduce network noise,we have presented a model(NERLS)based on Network Enhancement(NE)and Regularized Least Squares.First,we use the mean method to construct the virus and receptor similarity network based on the virus genome sequence similarity,receptor sequence similarity and known virus-receptor interactions.Secondly,the original edge weights in the similarity network are modified by NE.Finally,the regularized least squares algorithm is applied to predict virus-receptor interactions.The experimental results show that NERLS outperforms IILLS,DWNN,WP and Lap RLS.(3)Considering the fact that the biological data have missing values,a model(Pre VRIs)has been proposed based on missing value processing and matrix completion.Pre VRIs uses Gaussian Radial basis functions(GRB)to recalculate and fill in missing values in the viral protein sequence similarity,the viral genome sequence similarity,and the protein interaction network similarity.Then,the similarity kernel learning method is used to fuse two virus and receptor similarities respectively.Finally,the matrix completion method is used to predict the virus-receptor interactions.Experimental results show that Pre VRIs can effectively reduces the influence of missing values on predictive performance.(4)Different similarities can characterize different biological features from different perspectives.Based on multiple similarities and Linear Optimization of Matrix Completion,a model(LOMCVRI)has been proposed to predict Virus-Receptor Interactions.LOMCVRI utilizes GRB to recalculate and update missing values of the viral protein secondary structure similarity,the viral protein sequence similarity,and the viral genome sequence similarity.Then,updated viral similarities are fused into a comprehensive viral similarity by the mean method.Similarly,the receptor domain sequence similarity,receptor protein network similarity and receptor sequence similarity are also integrated into a comprehensive receptor similarity by the same method.Next,LOMCVRI uses K Nearest Neighbor(KNN)to fill zero rows and columns in the virus-receptor interaction matrix.The linear optimization method is used to predict potential virusreceptor interactions.The experimental results show that LOMCVRI can effectively integrate virus and receptor similarities,and is better than IILLS,KATZ,BRWH and NBI.(5)Considering that the linear optimization method only uses contribution of all neighbors of node i to node j and ignores contribution of all neighbors of node j to node i on the network link,a new method(MVILO)has been proposed via Improved Linear Optimization and Multiple View learning.MVILO uses multi-view learning method to fuse ”initial values”and ”updated values” of similarities into improved similarities.And then,these improved similarities are integrated into the final virus and receptor similarity network by the mean method,respectively.Finally,the improved linear optimization method is proposed to predict virus-receptor interactions.Experimental results show that MVILO is better than LOMCVRI,Pre VRIs,NERLS and IILLS.